CN109979194A - Heavy vehicle ramp based on Markov chain is creeped Activity recognition method - Google Patents

Heavy vehicle ramp based on Markov chain is creeped Activity recognition method Download PDF

Info

Publication number
CN109979194A
CN109979194A CN201910171316.XA CN201910171316A CN109979194A CN 109979194 A CN109979194 A CN 109979194A CN 201910171316 A CN201910171316 A CN 201910171316A CN 109979194 A CN109979194 A CN 109979194A
Authority
CN
China
Prior art keywords
data
height
vehicle
obtains
training set
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910171316.XA
Other languages
Chinese (zh)
Other versions
CN109979194B (en
Inventor
唐蕾
贾景池
马婧瑜
段宗涛
徐国强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Changan University
Original Assignee
Changan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Changan University filed Critical Changan University
Priority to CN201910171316.XA priority Critical patent/CN109979194B/en
Publication of CN109979194A publication Critical patent/CN109979194A/en
Application granted granted Critical
Publication of CN109979194B publication Critical patent/CN109979194B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Analytical Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Optimization (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention belongs to Activity recognition fields, and in particular to a kind of heavy vehicle ramp based on Markov chain is creeped Activity recognition method.The present invention is extracted by space-time characteristic using the GNSS running data of extensive heavy-duty commercial vehicle as data source and establishes vehicle behavior model with pattern learning, to recognize and predict that the ramp of heavy-duty commercial vehicle is creeped behavior.Time slice is carried out to data first, extracts the Time Continuous sample that car speed under same geographical location is zero, optimizes sample data precision with Kalman filtering, further promotes sample quality.Then the feature distribution of Markov Chain Monte Carlo Construction of A Model height of car difference is used, and carries out parameter Estimation using Metropolis Hastings algorithm, determines the behavioural characteristic of creeping of heavy-duty commercial vehicle.Finally, establishing the vehicle behavior model HVMove of logic-based recurrence.Reach the ramp driving state for effectively recognizing heavy-duty commercial vehicle, and accurately predicts the purpose of its ramp driving state.

Description

Heavy vehicle ramp based on Markov chain is creeped Activity recognition method
Technical field
The invention belongs to Activity recognition fields, and in particular to a kind of heavy vehicle ramp based on Markov chain is creeped row For recognition methods.
Background technique
Worldwide rapid economic growth increases considerably highway freight, a big chunk total transport cost be by What the high utilization rate of corresponding heavy vehicle was driven.Therefore, movement and behavior of the heavy vehicle on road are accurately identified, is The primary basis for reducing traffic accident, environmental impact assessment, reducing oil consumption, optimizing traffic department.Heavy vehicle is on road Tracking is the key that development automation heavy vehicle and advanced driving assistance system with Activity recognition, passes through identification heavy vehicle Movement and behavior, it will be appreciated that the dynamic characteristic of vehicle in road grade realizes accurately identifying for heavy goods vehicles motion state.Wherein The vehicle movement information that the vehicle space time information and behavioural analysis obtained based on track following is obtained, can predict possible driving Risk generates safety movement, avoids the generation of road accident.
In terms of the track following of heavy vehicle, the data acquisition in previous research mostly uses following methods:
1, video means acquisition method, this method mainly for exterior vehicle behavior identification, be easy by bad border naturally, The influence of the objective factors such as more vehicle real-time performance of tracking and Vehicular vibration impact, it is difficult to obtain the track of vehicle for being located at check frequency Data；
2, the acquisition method based on non-video means, it is main that vehicle behavior data, the calculation of use are obtained by multisensor Method places one's entire reliance upon the use of onboard sensor, and such as video camera, laser radar and radar, and this method is only applicable to single The Activity recognition of vehicle, further, since signal noise, the reasons such as vehicle driving regional vision is narrow, barrier is sudden, heavy The acquisition of the tracking data of vehicle is further limited.
In terms of the Activity recognition of creeping of heavy vehicle, existing research method has:
1, video data analysis method carries out convergence analysis processing by the data obtained to video means, from reality It is extracted in the temporal information of track data and clusters common condition sequence, so that different vehicle behaviors is identified, however, this method It is based on discrete event, data acquisition is sparse insufficient, and because without accurately considering the movement upgrading of heavy vehicle and dropping Grade feature and the high relationship of time complexity also limit heavy vehicle Activity recognition so that this method implementation effect is very poor Performance；
2, deep learning method combines simpler vehicle behavior identification model, and data are transmitted from one layer More complicated model is constructed to another layer, by the training automatically derived final Activity recognition model of mass data, to infer Other data not yet modeled lack versatility although depth learning technology can be improved the accuracy of vehicle behavior identification, Its application is limited by Deta sparseness, can not verify the accuracy of model, for example, when track data is imperfect, vehicle Complete driving behavior be unpredictable.In addition, the accuracy of identification of deep learning method is influenced by vehicle driving milimeter number, This makes application of this method in long range heavy-duty vehicle motion recognition tasks, and there are problems.
Summary of the invention
For the particularity on the in-orbit mark of heavy type commercial vehicle existing in the prior art and behavior, its traveling Activity recognition Precision is influenced by the problems such as VMT Vehicle-Miles of Travel, time complexity and universality, will cause the reduction of its recognition accuracy The problem of, the present invention provides one kind to be creeped Activity recognition method based on markovian heavy goods vehicles ramp, using following skill Art scheme is realized:
Heavy vehicle ramp based on Markov chain is creeped Activity recognition method, is included the following steps:
Step 1: the running data of heavy vehicle is obtained, data cleansing is carried out to running data, obtains sample running data, The sample running data includes position data, height observation data, vehicle speed data and time data；
Step 2: true map match being carried out to the sample running data that step 1 obtains and the date is segmented, and will be after segmentation Sample running data be divided into training set and test set, establish Height Prediction model, the height of training set observation data are defeated Enter Height Prediction model, obtains the optimum prediction value of the altitude information of training set；
Step 3: obtaining step 2 obtains the mistake between the optimum prediction value of the altitude information of training set and height observation data Poor threshold value, according to error threshold mark training set altitude information optimum prediction value obtain reference data, by reference data with The optimum prediction value of the altitude information for the training set that step 2 obtains is matched, and absolute altitude track is obtained；
Step 4: feature extraction being carried out to the absolute altitude track that step 3 obtains, establishes and is based on markovian heavy vehicle Climbing behavior model, the sample running data input heavy vehicle climbing behavior model for the test set that step 2 is obtained, is identified The behavior of creeping of test set medium-weight vehicle.
Further, the data cleansing in step 1 includes:
(1) for data departure time, receiving time and the identical duplicate running data of any item of longitude and latitude It is rejected；
(2) four are accurate to after decimal point to Time Continuous, longitude and latitude difference identical height value but occur obvious abnormal Running data carries out intermediate value processing；
(3) to the running data of missing, the method for taking front and back value to do mean value carries out interpolation processing；
(4) rearrangement processing is carried out to the running data of timing confusion.
Further, step 2 includes following sub-step:
Step 2.1: the sample running data that step 1 is obtained carry out true map match and using semantic segmentation method into The segmentation of row date, is divided into one section for the running data with similar features, and the sample running data after segmentation is divided into instruction Practice collection and test set；
Step 2.2: Height Prediction model is established using formula I:
H (k | k-1)=P (k | k-1)+Kg (k) [Y (k)-H (k | k-1)] formula I
Wherein, H is the altitude data of vehicle driving, and Kg (k) is kalman gain, and Y (k) is height observing matrix, H (k | k-1) is the optimum prediction value of k moment altitude information after Kalman filtering, P (k | k-1) be after Kalman filtering the k moment it is high The covariance matrix of degree evidence, k-1 moment are the observation moment, and the k moment is prediction time；
Step 2.3: the height observation data of training set being inputted into Height Prediction model, obtain the altitude information of training set Optimum prediction value.
Further, step 3 includes following sub-step:
Step 3.1: obtaining step 2 obtains the optimum prediction value and height observation data of the altitude information of training set, utilizes Formula II obtains error threshold AT between the two,
Wherein, n is the time hop counts that segmentation obtains, SSPKFor the short stay point at k moment；
Step 3.2: the error threshold AT obtained according to step 3.1 carries out the optimum prediction value of the altitude information of training set Label, the altitude information difference if the two neighboring moment are labeled as 1 if being greater than error threshold AT, 0 are otherwise labeled as, after label The optimum prediction value of the altitude information of training set is denoted as reference data；
Step 3.3: by the optimal pre- of the altitude information of the reference data that step 3.2 obtains and the training set that step 2 obtains Measured value is matched, and absolute altitude track is obtained.
Further, step 4 includes following sub-step:
Step 4.1: feature extraction being carried out to the absolute altitude track that step 3 obtains, the feature is difference in height and continuous height Difference and；
Step 4.2: it is established using formula III based on markovian heavy vehicle climbing behavior model,
P (D | HD) indicate that the probability that car travel mode is judged by difference in height, D indicate that driving mode, D=0 indicate Level road, D=1 indicate to go up a slope, and HD indicates difference in height, and α and β are Optimal Parameters；
Step 4.3: the sample running data input heavy vehicle climbing behavior model for the test set that step 2 is obtained is known Not Chu test set medium-weight vehicle behavior of creeping.
Further, in step 4.2 Optimal Parameters α and β selection specifically:
(1) prior probability distribution of Gaussian Profile constructing variable α and β is used；
(2) probability P (HD that creeps based on Bernoulli Jacob's variable is establishedi) and parameter alpha and β mapping relations；
(3) priori Gaussian Profile N (μ, σ are used2) construction target distribution be P (D | HD) markov chain, to determine between state Transition probability, generate heavy vehicle climbing behavior model probability distribution sample；
(4) extension MCMC is sampled α and β from two Gaussian Profiles, obtains the sample set of each parameter, in turn The average value and maximum likelihood state for obtaining all samples, finally determine reasonable parameter alpha and β.
The invention has the following beneficial effects:
(1) present invention uses extensive heavy-duty commercial vehicle running data, is extracted by space-time characteristic and is established with pattern learning Vehicle behavior model, wherein using the climbing behavioural characteristic of MCMC model analysis and selection vehicle, and extend Metropolis Hastings carries out parameter Estimation, to reach the ramp driving state for effectively recognizing heavy-duty commercial vehicle, and accurately predicts Its ramp driving state.
(2) present invention extends Markov chain Monte-Carlo (Markov Chain Monte Carlo, MCMC) method Parameter Estimation, the behavior model of creeping of Lai Jianli heavy type commercial vehicle are carried out to model；Simultaneously in view of heavy type commercial vehicle exists Particularity, accuracy of identification in track and behavior can be influenced by VMT Vehicle-Miles of Travel, time complexity and the problems such as universality, The extensive running data for obtaining heavy-duty commercial vehicle by position sensor is proposed, is creeped behavior for its typical long ramp It is recognized and is predicted.Have greatly improved in heavy vehicle Activity recognition precision aspect of creeping, model can be effectively reduced The ramp driving state of heavy-duty commercial vehicle can be more accurately predicted in the loss of complexity and model performance.
Detailed description of the invention
Fig. 1 is that car ramp provided by the invention is creeped the flow chart of Activity recognition method；
Fig. 2 is the track on practical map to be highly mark；
Fig. 3 is influence of the data sampling point to identification probability.
Specific embodiment
The technical term occurred in patent is explained first:
Markov chain Monte-Carlo (Markov Chain Monte Carlo, MCMC) method: this method is by Ma Erke Husband (Markov) process is introduced into Monte Carlo (Monte Carlo) simulation, realizes that sampling distribution changes with the progress of simulation The dynamic analog of change, compensate for traditional monte carlo integration can only static simulation defect.MCMC is a kind of simple and effective Calculation method, in many fields to being widely applied, such as statistics object, Bayes (Bayes) problem, computer problem.
Kalman filtering: Kalman filtering utilizes linear system state equation, observation data is inputted by system, to system State carries out optimal estimation.
Absolute altitude track: the track as obtained from marking the height of each data on practical map.
Reference data: an error threshold of the same longitude and latitude height measurements of GNSS is obtained by linear regression.
GNSS running data: the vehicle positioned by heavy vehicle self-sensor device combining global navigational satellite system Running data, including data departure time, data receipt time, GNSS longitude and latitude, height, speed, telltale mark, north and south latitude mark Note, thing are labeled.
The track GNSS: by inventing the track data obtained after the data processing.
HVMove model: heavy vehicle ramp established by the present invention is creeped Activity recognition model.
Heavy vehicle ramp based on Markov chain is creeped Activity recognition method, is included the following steps:
Step 1: the running data of heavy vehicle is obtained, data cleansing is carried out to running data, obtains sample running data, The sample running data includes position data, height observation data, vehicle speed data and time data；
Step 2: true map match being carried out to the sample running data that step 1 obtains and the date is segmented, and will be after segmentation Sample running data be divided into training set and test set, establish Height Prediction model, the height of training set observation data are defeated Enter Height Prediction model, obtains the optimum prediction value of the altitude information of training set；
Step 3: obtaining step 2 obtains the mistake between the optimum prediction value of the altitude information of training set and height observation data Poor threshold value, according to error threshold mark training set altitude information optimum prediction value obtain reference data, by reference data with The optimum prediction value of the altitude information for the training set that step 2 obtains is matched, and absolute altitude track is obtained；
Step 4: feature extraction being carried out to the absolute altitude track that step 3 obtains, establishes and is based on markovian heavy vehicle Climbing behavior model, the sample running data input heavy vehicle climbing behavior model for the test set that step 2 is obtained, is identified The behavior of creeping of test set medium-weight vehicle.
The present invention passes through space-time characteristic extraction and mode using the GNSS running data of extensive heavy-duty commercial vehicle as data source Vehicle behavior model is established in study, to recognize and predict that the ramp of heavy-duty commercial vehicle is creeped behavior.
Preferably, the running data for heavy vehicle being obtained in step 1 is a certain heavy-duty commercial vehicle positioned using GNSS Running data in the data cover vehicle 9 days, saves in Shaanxi and the domestic Beijing-Kunming fastlink in Shanxi Province, G108 national highway, S331 The driving trace in road, totally 57698 data.
Specifically, the data cleansing in step 1 includes:
(1) for data departure time, receiving time and the identical duplicate running data of any item of longitude and latitude It is rejected；
(2) four are accurate to after decimal point to Time Continuous, longitude and latitude difference identical height value but occur obvious abnormal Running data carries out intermediate value processing；
(3) to the running data of missing, the method for taking front and back value to do mean value carries out interpolation processing；, such as in acquisition Lack in one running data ride-height value this, using the height value and latter running data of previous running data The mean value of data height value is as filling；
(4) rearrangement processing is carried out to the running data of timing confusion；Such as in sampling, due to by geographical ring The influence in border, signal strength, so that putting in order for running data is not necessarily to time sequencing.
It being operated by above data cleansing, the data that can make are more accurate, rationally, convenient post-processing.
Specifically, step 2 includes following sub-step:
Step 2.1: the sample running data that step 1 is obtained carry out true map match and using semantic segmentation method into The segmentation of row date, is divided into one section for the running data with similar features, and the sample running data after segmentation is divided into instruction Practice collection and test set；
Preferably, the true map match is that running data is projected on true map with ArcGis, to data Point is corrected and is analyzed, and to be corrected and be analyzed to data point, determines travel route of the heavy-duty commercial vehicle on map.
Preferably, the semantic segmentation indicates to believe data with similar features such as position, height, speed and time etc. Breath is classified.
Preferably, the sample running data after segmentation is divided into training set and test set according to 7 to 3 ratio, i.e., 70% Data for training, remaining 30% for testing.
Preferably, the date is divided into 24 hours.
Step 2.2: Height Prediction model is established using formula I:
H (k | k-1)=P (k | k-1)+Kg (k) [Y (k)-H (k | k-1)] formula I
Wherein, H is the altitude data of vehicle driving, and Kg (k) is kalman gain, and Y (k) is height observing matrix, H (k | k-1) is the optimum prediction value of k moment altitude information after Kalman filtering, P (k | k-1) be after Kalman filtering the k moment it is high The covariance matrix of degree evidence, k-1 moment are the observation moment, and the k moment is prediction time；
Specifically, first with Kalman filtering, it is high by the available k moment state of the altitude information of k-1 moment state The predicted value of degree evidence；
H (k)=ψ H (k-1)+γ W (k) formula 1
Wherein, for k-th of state at a certain moment, H is the altitude data of vehicle driving, and H (k-1) represents current The data that the vehicle sample training of state input is concentrated, H (k) represent the data that the vehicle sample training of next state is concentrated, W It (k) is process noise matrix, γ=[1] and ψ=[1] are state transition matrix.
Secondly, giving the observation model of altitude information, available k moment state altitude information in k-th of state Observation；
Y (k)=NG (k)+D (k) formula 2
Wherein Y (k) is height observing matrix, and N=[1] is observing matrix, and G (k) is true height matrix, and D (k) is to see The noise matrix measured, D (k) are observation noise matrix.
The covariance matrix P of the altitude information at current k-1 moment is updated, kalman gain Kg is calculated and determines parameter ψ, wherein Q is the covariance of process noise:
P (k | k-1)=ψ P (k-1 | k-1) ψT+ Q (k-1) formula 3
Wherein, P (k | k-1) is the covariance matrix of the k moment state altitude information after Kalman filtering, and the form is The canonical form meaning of Kalman filtering is to ask the k moment by the k-1 moment.
Predicted value and observation can be connected simultaneously and (find out to come respectively by them, then pass through construction karr Graceful gain distributes the weight of predicted value and observation), the observation and observation of NextState are determined with this, by prediction section It is divided to and observes after the Gaussian Profile of part two is multiplied to still Gaussian distributed, which relatively determines using two-part weight As a result more accurate；
It is as follows come the weight for distributing predicted value and observation to construct kalman gain Kg:
Kg (k)=P (k-1 | k-1) NT[NP(k|k-1)NT+R(k-1)]-1Formula 4
Wherein, R (k) is the covariance matrix of D (k), the observation that the predicted value obtained according to formula 1 and formula 2 obtain, when k The optimal height value at quarter can be obtained by the weight that formula 3 and formula 4 calculate.
Step 2.3: the height observation data of training set being inputted into Height Prediction model, obtain the altitude information of training set Optimum prediction value.
The above method has shortage lag, state vector richer using Kalman filtering relative to mean value and median filter Richness is handling the advantages such as roughness point, carries out the processing of heavy truck track data, and several filtering modes are more accurate before obtaining relatively, Better effect.
The purpose of step 2 is: random noise problem of the sample data obtained for step 1 in elevation dimension, to sample Notebook data is filtered and is segmented, the precision and quality of Lai Tisheng sample data.
Specifically, step 3 includes following sub-step:
Step 3.1: obtaining step 2 obtains the optimum prediction value and height observation data of the altitude information of training set, utilizes Formula II obtains error threshold AT between the two,
Wherein, n is the time hop counts that segmentation obtains, SSPKFor the short stay point at k moment, short stay point SSP is represented One section of geodata that heavy vehicle is travelled with 0m/s speed in a certain time interval；
Preferably, linear regression is chosen to be shown below as the model of error threshold:
hn(t)=θ0+θ1T formula 11
Wherein hn(t) function of the height about time series, θ are indicated0Indicate intercept, θ1Indicate slope, t is sample data Time.
Step 3.2: the error threshold AT obtained according to step 3.1 carries out the optimum prediction value of the altitude information of training set Label, the altitude information difference if the two neighboring moment are labeled as 1 if being greater than error threshold AT, 0 are otherwise labeled as, after label The optimum prediction value of the altitude information of training set is denoted as reference data；
Step 3.3: by the optimal pre- of the altitude information of the reference data that step 3.2 obtains and the training set that step 2 obtains Measured value is matched, and obtains absolute altitude track, as shown in Figure 2.
Specifically, step 4 includes following sub-step:
Step 4.1: feature extraction being carried out to the absolute altitude track that step 3 obtains, the feature is difference in height and continuous height Difference and；
Wherein, difference in height can reflect the opposite number to climb with decline of heavy truck each track point height in continuous time period Value, therefore, two tracing points continuous to the time calculate its difference in height, obtain difference in height feature；
In addition, continuous difference in height and being able to reflect vehicle duration under same driving mode.This feature is advantageous Track and height track are continuously driven in one for showing heavy-duty commercial vehicle.
Step 4.2: heavy vehicle climbing behavior model is established using formula III,
P (D | HD) indicate that the probability that car travel mode is judged by difference in height, D indicate that driving mode, D=0 indicate Level road, D=1 indicate to go up a slope, and HD indicates difference in height, and α and β are Optimal Parameters；
Step 4.3: the sample running data input heavy vehicle climbing behavior model for the test set that step 2 is obtained is known Not Chu test set medium-weight vehicle behavior of creeping.
Specifically, in step 4.2 Optimal Parameters α and β selection specifically:
(1) prior probability distribution of Gaussian Profile constructing variable α and β is used；When initialization, we are prior probability distribution Two parameter μs are chosen with Gaussian Profile, τ chooses μ=0 and τ=0.05 and establishes Gaussian Profile, and extension MCMC methodology adjusts μ, τ, To estimate the optimal α and β of HVMove model.
(2) probability P (HD that creeps based on Bernoulli Jacob's variable is establishedi) and parameter alpha and β mapping relations；In HVMove model In, we simulate level road and climbing as Bernoulli Jacob's variable, P (HDi) probabilistic model is converted to the Bernoulli Jacob of only 01 variable Variable Ber (P (HDi)), 0 represents level road, and 1 represents climbing, and i is height difference sequence, correction model are as follows:
(3) priori Gaussian Profile N (μ, σ are used2) construction target distribution be P (D | HD) markov chain, to determine between state Transition probability, generate HVMove model probability distribution sample；
New state Y is provided from Gaussian Profile first, calculates the receptance of the state Independent random variable u is obtained from being uniformly distributed in U simultaneously, by comparing u and a, obtains new state composition and meet to be distributed as P (D | HD) markov chain generate the probability distribution sample of HVMove model to determine the transition probability between state.
(4) extension MCMC is sampled α and β from two Gaussian Profiles, obtains the sample set of each parameter, in turn The average value and maximum likelihood state for obtaining all samples determine reasonable parameter alpha and β finally to optimize HVMove model.
In each iteration, α and β need to be estimated according to previous state, if the parameter chosen meets real data point Cloth then receives current state, otherwise then refuses current state.Therefore, it can get the sample of each parameter under particular iteration number This set, and then the average value and maximum likelihood state of all samples are obtained, determine reasonable parameter alpha and β finally to optimize HVMove model.
The following provides a specific embodiment of the present invention, it should be noted that the invention is not limited to implement in detail below Example, all equivalent transformations made on the basis of the technical solutions of the present application each fall within protection scope of the present invention.
Embodiment
The foundation of HVMove model is carried out using 2000 sample data test set sampled points first.In view of sampled point pair The influence of Activity recognition probability, we compared the Activity recognition probability under different data scale, as shown in Figure 3.It is respectively adopted 500,1000,2000,3000,4000,5000 data points of sampling carry out the foundation of HVMove model.We have chosen continuously Corresponding three tracing points of difference in height and respectively 5.5m, 6.0m, 6.5m are used to compare, it can be observed that when data scale reaches When 4000, identification probability has reached a stable state, will not be influenced again by sampling scale.It is continuous high to three at this time The Activity recognition probabilistic forecasting of creeping for spending poor sum is respectively 32.50%, 99.37%, 100%.

Claims (6)

  1. The Activity recognition method 1. the heavy vehicle ramp based on Markov chain is creeped, which comprises the steps of:
    Step 1: the running data of heavy vehicle is obtained, data cleansing is carried out to running data, obtains sample running data, it is described Sample running data includes position data, height observation data, vehicle speed data and time data；
    Step 2: the sample running data obtained to step 1 carries out true map match and the date is segmented, and by the sample after segmentation This running data is divided into training set and test set, establishes Height Prediction model, and the height observation data input of training set is high Prediction model is spent, the optimum prediction value of the altitude information of training set is obtained；
    Step 3: obtaining step 2 obtains the error threshold between the optimum prediction value of the altitude information of training set and height observation data Value marks the optimum prediction value of the altitude information of training set to obtain reference data, by reference data and step 2 according to error threshold The optimum prediction value of the altitude information of obtained training set is matched, and absolute altitude track is obtained；
    Step 4: feature extraction being carried out to the absolute altitude track that step 3 obtains, establishes and is climbed based on markovian heavy vehicle Behavior model, the sample running data input heavy vehicle climbing behavior model for the test set that step 2 is obtained, identifies test Collect the behavior of creeping of medium-weight vehicle.
  2. 2. the heavy vehicle ramp based on Markov chain is creeped Activity recognition method as described in claim 1, feature exists In the data cleansing in step 1 includes:
    (1) data departure time, receiving time and the identical duplicate running data of any item of longitude and latitude are carried out It rejects；
    (2) four are accurate to after decimal point to Time Continuous, longitude and latitude difference and identical the obvious abnormal travelings of height value but occur Data carry out intermediate value processing；
    (3) to the running data of missing, the method for taking front and back value to do mean value carries out interpolation processing；
    (4) rearrangement processing is carried out to the running data of timing confusion.
  3. 3. the heavy vehicle ramp based on Markov chain is creeped Activity recognition method as described in claim 1, feature exists In step 2 includes following sub-step:
    Step 2.1: true map match being carried out to the sample running data that step 1 obtains and day is carried out using semantic segmentation method Phase segmentation, is divided into one section for the running data with similar features, and the sample running data after segmentation is divided into training set And test set；
    Step 2.2: Height Prediction model is established using formula I:
    H (k | k-1)=P (k | k-1)+Kg (k) [Y (k)-H (k | k-1)] formula I
    Wherein, H be vehicle driving altitude data, Kg (k) be kalman gain, Y (k) be height observing matrix, H (k | It k-1 is) the optimum prediction value of k moment altitude information after Kalman filtering, and P (k | k-1) it is k moment high degree after Kalman filtering According to covariance matrix, the k-1 moment be observation the moment, the k moment be prediction time；
    Step 2.3: the height observation data of training set being inputted into Height Prediction model, obtain the optimal of the altitude information of training set Predicted value.
  4. 4. the heavy vehicle ramp based on Markov chain is creeped Activity recognition method as described in claim 1, feature exists In step 3 includes following sub-step:
    Step 3.1: obtaining step 2 obtains the optimum prediction value and height observation data of the altitude information of training set, utilizes formula II Error threshold AT between the two is obtained,
    Wherein, n is the time hop counts that segmentation obtains, SSPKFor the short stay point at k moment；
    Step 3.2: the error threshold AT obtained according to step 3.1 marks the optimum prediction value of the altitude information of training set Note, the altitude information difference if the two neighboring moment are labeled as 1 if being greater than error threshold AT, 0 are otherwise labeled as, by the instruction after label The optimum prediction value for practicing the altitude information of collection is denoted as reference data；
    Step 3.3: by the optimum prediction value of the reference data that step 3.2 obtains and the altitude information for the training set that step 2 obtains It is matched, obtains absolute altitude track.
  5. 5. the heavy vehicle ramp based on Markov chain is creeped Activity recognition method as described in claim 1, feature exists In step 4 includes following sub-step:
    Step 4.1: feature extraction carried out to the absolute altitude track that step 3 obtains, the feature be difference in height and continuous difference in height and；
    Step 4.2: it is established using formula III based on markovian heavy vehicle climbing behavior model,
    P (D | HD) indicate that the probability that car travel mode is judged by difference in height, D indicate that driving mode, D=0 indicate level road, D=1 indicates to go up a slope, and HD indicates difference in height, and α and β are Optimal Parameters；
    Step 4.3: the sample running data input heavy vehicle climbing behavior model for the test set that step 2 is obtained identifies The behavior of creeping of test set medium-weight vehicle.
  6. 6. the heavy vehicle ramp based on Markov chain is creeped Activity recognition method as claimed in claim 5, feature exists In the selection of Optimal Parameters α and β in step 4.2 specifically:
    (1) prior probability distribution of Gaussian Profile constructing variable α and β is used；
    (2) probability P (HD that creeps based on Bernoulli Jacob's variable is establishedi) and parameter alpha and β mapping relations；
    (3) priori Gaussian Profile N (μ, σ are used2) construction target distribution be P (D | HD) markov chain, to determine the transfer between state Probability generates the probability distribution sample of heavy vehicle climbing behavior model；
    (4) extension MCMC is sampled α and β from two Gaussian Profiles, obtains the sample set of each parameter, and then obtain The average value and maximum likelihood state of all samples, finally determine reasonable parameter alpha and β.
CN201910171316.XA 2019-03-07 2019-03-07 Heavy vehicle ramp crawling behavior identification method based on Markov chain Active CN109979194B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910171316.XA CN109979194B (en) 2019-03-07 2019-03-07 Heavy vehicle ramp crawling behavior identification method based on Markov chain

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910171316.XA CN109979194B (en) 2019-03-07 2019-03-07 Heavy vehicle ramp crawling behavior identification method based on Markov chain

Publications (2)

Publication Number Publication Date
CN109979194A true CN109979194A (en) 2019-07-05
CN109979194B CN109979194B (en) 2020-07-10

Family

ID=67078072

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910171316.XA Active CN109979194B (en) 2019-03-07 2019-03-07 Heavy vehicle ramp crawling behavior identification method based on Markov chain

Country Status (1)

Country Link
CN (1) CN109979194B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112906742A (en) * 2021-01-19 2021-06-04 重庆邮电大学 Two-wheel vehicle identification system and method based on 5G + V2X mobile terminal and high-precision map
CN113465613A (en) * 2021-07-22 2021-10-01 全图通位置网络有限公司 Map matching optimization method for tunnel network positioning in urban rail transit
WO2021248655A1 (en) * 2020-06-11 2021-12-16 西安电子科技大学 Multi-geomagnetic sensor speed measurement system and speed measurement method therefor

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102568200A (en) * 2011-12-21 2012-07-11 辽宁师范大学 Method for judging vehicle driving states in real time
JP2012208701A (en) * 2011-03-29 2012-10-25 Fuji Heavy Ind Ltd Driving support system for vehicle
CN107967486A (en) * 2017-11-17 2018-04-27 江苏大学 A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models
DE102018120677A1 (en) * 2017-08-28 2019-02-28 Ford Global Technologies, Llc Systems and methods for adjusting the uniformity of vehicle speed control

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012208701A (en) * 2011-03-29 2012-10-25 Fuji Heavy Ind Ltd Driving support system for vehicle
CN102568200A (en) * 2011-12-21 2012-07-11 辽宁师范大学 Method for judging vehicle driving states in real time
DE102018120677A1 (en) * 2017-08-28 2019-02-28 Ford Global Technologies, Llc Systems and methods for adjusting the uniformity of vehicle speed control
CN107967486A (en) * 2017-11-17 2018-04-27 江苏大学 A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021248655A1 (en) * 2020-06-11 2021-12-16 西安电子科技大学 Multi-geomagnetic sensor speed measurement system and speed measurement method therefor
US11908319B2 (en) 2020-06-11 2024-02-20 Xidian University Multi-geomagnetic sensor speed measurement system and a speed measurement method using the same
CN112906742A (en) * 2021-01-19 2021-06-04 重庆邮电大学 Two-wheel vehicle identification system and method based on 5G + V2X mobile terminal and high-precision map
CN113465613A (en) * 2021-07-22 2021-10-01 全图通位置网络有限公司 Map matching optimization method for tunnel network positioning in urban rail transit
CN113465613B (en) * 2021-07-22 2023-12-26 全图通位置网络有限公司 Map matching optimization method for tunnel network positioning in urban rail transit

Also Published As

Publication number Publication date
CN109979194B (en) 2020-07-10

Similar Documents

Publication Publication Date Title
CN108492564B (en) Method and system for measuring running speed of highway vehicle based on road network matching
CN109946731B (en) Vehicle high-reliability fusion positioning method based on fuzzy self-adaptive unscented Kalman filtering
CN110168520A (en) A kind of intelligence road traffic method for detecting abnormality
CN101576384B (en) Indoor movable robot real-time navigation method based on visual information correction
Ghaleb et al. Improved vehicle positioning algorithm using enhanced innovation-based adaptive Kalman filter
JP5162849B2 (en) Fixed point position recorder
CN109979194A (en) Heavy vehicle ramp based on Markov chain is creeped Activity recognition method
CN103336863A (en) Radar flight path observation data-based flight intention recognition method
CN111862165A (en) Target tracking method for updating Kalman filter based on deep reinforcement learning
CN112462401B (en) Urban canyon rapid detection method and device based on floating vehicle track data
CN103678925A (en) Flight path classification method based on auxiliary information source
Liu et al. Determination of routing velocity with GPS floating car data and webGIS-based instantaneous traffic information dissemination
CN115638767B (en) Ground subsidence monitoring method
CN117029840A (en) Mobile vehicle positioning method and system
CN115630056B (en) Road inspection quantitative assessment system and method based on GPS track positioning
CN101388000A (en) Method for evaluating guide marker spatial arrangement based on view superpose analysis
CN111524358A (en) Regional radiation sexual communication flow prediction method
CN114565176B (en) Long-term ship track prediction method
CN109379716A (en) A kind of indoor orientation method and system for safety monitoring project
Murphy et al. Road-assisted multiple target tracking in clutter
Zhang et al. The research for a kind of information fusion model based on BP neural network with multi position sources and big data selection
Chen et al. Research on evaluation and prediction method of link travel time based on floating car data by simulation
Liu et al. A YOLO-X Deep Learning Approach to Detect Traffic Targets from UAV Video with On-Board Vehicle Data Validation
CN116647651B (en) Unmanned aerial vehicle construction monitoring method and system based on Beidou satellite
CN117407477B (en) Geographic information data evolution recognition processing method, system and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant